⚡ electrical engineering

A 850 µm 2 low-power Smart Temperature Sensor for IoT Applications

This paper presents a fully integrated, low-power smart temperature sensor fabricated in 65-nm CMOS that achieves an area of 850 µm², consumes 5.5 µW, and delivers high resolution and accuracy for IoT thermal monitoring by utilizing a self-clocked, reference-free ring oscillator-based sensing methodology.

Original authors: Neethu Johny, Jayagowri R

Published 2026-06-24

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Neethu Johny, Jayagowri R

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you have a tiny, battery-powered computer chip that needs to know how hot it is to avoid overheating, but it can't afford to carry a heavy, expensive, or power-hungry thermometer. That's the problem this paper solves.

The authors, Neethu Johny and R. Jayagowri, have built a smart temperature sensor that is incredibly small (about the size of a grain of sand, specifically 850 square micrometers) and uses almost no electricity. It's designed specifically for the "Internet of Things" (IoT)—those tiny devices like smart watches, environmental monitors, or sensors in a factory that run on small batteries for years.

Here is how their invention works, explained with everyday analogies:

1. The Problem: The "Heavy Backpack"

Traditional temperature sensors are like carrying a heavy backpack. They often need extra parts (like special resistors or external chips) to work, which takes up space and drains the battery. The authors wanted to build a sensor that is as light as a feather, using only the standard digital parts already inside the computer chip, without needing any extra "backpack" accessories.

2. The Core Idea: The "Two Runners" Race

Instead of using a thermometer that measures heat directly, this sensor uses a clever race between two runners.

The Runners: The sensor has two tiny electronic circuits called Ring Oscillators. Think of these as two runners running in circles. Every time they complete a lap, they "tick" a counter.
The Twist: The two runners are built slightly differently.
- Runner A (VCO1): Built with "short legs" (shorter transistors). This runner speeds up significantly when it gets hot.
- Runner B (VCO2): Built with "longer legs" (longer transistors). Thanks to a quirk in physics called the "Reverse Short-Channel Effect," this runner is built to be less sensitive to the heat. It speeds up, but not as dramatically as Runner A.

3. How It Measures Temperature: The "Stopwatch"

The sensor doesn't need an outside clock or a reference temperature. It runs its own race:

The Start: Both runners start running at the same time.
The Finish Line: The sensor sets a finish line for Runner B (the less sensitive one). Let's say Runner B needs to run exactly 1,000 laps to finish.
The Count: While Runner B is running those 1,000 laps, the sensor counts how many laps Runner A completed.
- If it's cold, Runner A is slow. By the time Runner B finishes 1,000 laps, Runner A might have only done 500 laps.
- If it's hot, Runner A is fast. By the time Runner B finishes 1,000 laps, Runner A might have done 800 laps.
The Result: The sensor simply looks at the number of laps Runner A did. That number is the temperature. No complex math or external tools are needed; it's just a digital count.

4. The "Near-Threshold" Trick: Running on Fumes

Most electronics need a strong battery charge (voltage) to run fast. However, to save power, this sensor runs on a very low voltage, close to the minimum amount needed to keep the transistors alive. This is called Near-Threshold Voltage operation.

Think of it like a car engine idling. It's barely running, but it's incredibly fuel-efficient. This allows the sensor to work for a very long time on a tiny battery, which is perfect for IoT devices that might be buried in a wall or stuck on a tree branch.

5. The Results: Small, Fast, and Accurate

The paper claims the following achievements for their design:

Size: It's tiny (850 µm²), fitting easily onto a chip without taking up much room.
Power: It uses very little energy (5.5 microwatts), which is like a tiny whisper of electricity.
Accuracy: It can tell the temperature difference as small as 0.06°C. After a simple one-time adjustment (calibration), it is accurate within about ±2°C across a wide range of temperatures (from freezing cold at -40°C to very hot at 125°C).
Efficiency: It converts temperature to a digital number so efficiently that it uses less energy per measurement than many other high-tech sensors on the market.

Summary

In short, the authors created a digital thermometer that doesn't need a battery pack or a reference tool. It works by timing a race between two slightly different electronic runners. Because it uses standard digital parts and runs on very low power, it is an ideal, lightweight solution for keeping track of heat in the billions of tiny, battery-powered devices that make up the Internet of Things.

Technical Summary: A 850 µm² Low-Power Smart Temperature Sensor for IoT Applications

Problem Statement
The increasing complexity and power density of System-on-Chip (SoC) designs have created significant thermal challenges, necessitating effective on-chip temperature monitoring to prevent performance degradation, accelerated aging, and permanent damage. While traditional discrete sensors (thermistors, thermocouples) offer high accuracy, their reliance on external components hinders system integration and increases cost. Conversely, existing on-chip solutions face trade-offs: resistor-based sensors require large areas and higher power; BJT-based sensors suffer from process sensitivity and poor scalability in modern CMOS; and many standard CMOS sensors struggle to operate reliably under the near-threshold voltage (NTV) regimes required for ultra-low-power Internet-of-Things (IoT) nodes. At NTV, small temperature variations cause disproportionate changes in transistor delay and leakage, making robust sensing a critical design constraint for energy-constrained applications.

Methodology
The authors propose a fully integrated, self-clocked, temperature-to-digital converter (TDC) based smart sensor designed for NTV operation. The architecture eliminates the need for external references or analog biasing circuits by relying exclusively on standard digital logic gates.

Sensing Principle: The core sensing mechanism utilizes two ring oscillators (VCO1 and VCO2) engineered to exhibit distinct temperature sensitivities.
- VCO1 uses standard short-channel transistors ( $L=200$ nm).
- VCO2 employs longer channel transistors ( $L=600$ nm).
- Reverse Short-Channel Effect (RSCE): The design exploits the RSCE inherent in planar CMOS processes. The longer channel length in VCO2 results in a lower effective threshold voltage ( $V_{TH}$ ) compared to VCO1. In the NTV region, this creates a specific interplay between mobility degradation (negative temperature coefficient) and threshold voltage reduction (positive temperature coefficient). VCO2 is designed to have a "flattened" frequency-temperature slope, acting as a reference, while VCO1 maintains a strong positive slope.
Quantization: The sensor digitizes the ratio of the oscillation frequencies ( $F_{VCO1}/F_{VCO2}$ $F_{V C O 1} / F_{V C O 2}$ ) using a Time-to-Digital Converter (TDC) scheme.
- Operation: Upon a start signal, both oscillators run. VCO2 drives a reference counter (Counter 2) set to a preset count ( $N=2^{10}$ ). When Counter 2 reaches this limit, it generates an End of Conversion (EOC) signal, halting both oscillators.
- Output: The count accumulated by Counter 1 (driven by VCO1) during this interval is latched as the digital output ( $D_{out}$ ). Since $D_{out} = N \times (f_{clk1}/f_{clk2})$ , the output is directly proportional to the frequency ratio, and thus to the temperature.
Implementation: The design is implemented in a 65-nm CMOS process with a supply voltage of 0.5 V. The sensor core includes 5-stage and 9-stage inverters for VCO1 and VCO2, respectively, alongside 13-bit and 10-bit asynchronous counters.

Key Contributions

Purely Digital Architecture: The sensor is realized entirely with standard digital logic gates, removing the need for analog components like band-gap references or voltage regulators, thereby enhancing scalability and reducing area.
RSCE-Based Threshold Engineering: Instead of using body biasing or multiple transistor flavors (which require extra masks or analog circuitry), the authors utilize the Reverse Short-Channel Effect via channel length scaling to create the necessary threshold voltage difference between the two oscillators.
Near-Threshold Optimization: All sub-systems are redesigned and sized specifically for reliable operation at 0.5 V (NTV), addressing the stability and timing challenges inherent to low-voltage IoT applications.
Self-Referenced Operation: The system operates without an external frequency source, using the internal frequency ratio for temperature determination.

Results
The proposed sensor was characterized through post-layout simulations and Monte Carlo analysis in a 65-nm process:

Area and Power: The sensor occupies a compact area of 850 µm² (0.00085 mm²). At 0.5 V, it consumes 5.5 µW of power (2.9 µW for digital logic, 2.6 µW for VCOs).
Performance Metrics:
- Resolution: 0.06°C.
- Temperature Range: -40°C to 125°C.
- Accuracy: After one-point calibration at 60°C with polynomial correction, the peak-to-peak inaccuracy is +1.8°C to -3.5°C (specifically noted as +1.6°C/-2.9°C in the conclusion for the typical corner).
- Energy Efficiency: The energy per conversion is 0.133 nJ, yielding a Resolution Figure of Merit (R-FoM) of 0.46 pJ·K².
Robustness: Monte Carlo simulations (1000 runs) and supply voltage sweeps (0.45V–0.55V) demonstrate stability against process variations and supply fluctuations. The frequency ratio exhibits Proportional-to-Absolute-Temperature (PTAT) behavior.

Significance
The paper positions this work as a viable solution for energy-limited IoT sensor nodes and wireless sensor networks where ultra-low-power operation and dense integration are paramount. By demonstrating that accurate thermal monitoring can be achieved using only standard digital logic in a near-threshold regime, the authors argue for a shift away from complex analog-heavy sensing paradigms. The use of RSCE for threshold engineering offers a cost-effective, mask-free method to tune sensor sensitivity. The results confirm that purely digital, mismatch-aware architectures can provide reliable thermal feedback for closed-loop adaptive voltage and frequency scaling, enabling robust, ultra-low-energy system-on-chip designs.